OpenAI's programming capabilities are expected to catch up with Claude within three months, with the core being that Codex has contributed more high-quality data and practices, which will be used to train, iterate, and fine tune the next generation models for acceleration. The Codex series models of GPT were trained together with the Codex framework, and many people underestimate the power of this statement. Currently, in order to maximize GPT's programming ability, it should be used together in Codex, rather than in Claude Code. If you have used Codex, you should be able to understand what I said. Codex's code engineering ability is already stronger than Claude's, but in terms of understanding product requirements, it is indeed insufficient. Therefore, I usually use Claude to sort out requirements first and then use Codex to develop. Can you explain in detail what models and frameworks are? They are not pieced together later, but designed together -The Codex model is not a universal model+code fine-tuning -It was trained together with the Codex framework -Meaning: the model understands the internal mechanisms of the framework, and the framework understands the output patterns of the model Result: Fewer misunderstandings, more accurate output, fewer iterations, and higher quality are the true understanding of code. The three-tier architecture of Codex 1. Model layer - providing core intelligence -OpenAI's flagship programming models (such as gpt-5.3-codex, GPT 5.4) -Perform structured reasoning before responding -Understand code logic, architecture design, and best practices 2. Framework layer (Harness) - connects to the real environment -An open-source execution framework that enables' execution 'rather than just' suggestion ' -Use compression and other techniques to manage contextual windows -Enable the model to truly manipulate files, run commands, and test code 3. Surfaces - Diversified Interaction Modes Codex App: Desktop application supporting parallel workflow CLI: Terminal and CI/CD Integration VS Code extension: iterating within the editor Mini: Lightweight remote task execution